virtualization and fault tolerance in cloud...

Virtualization and Fault Tolerance

in Cloud Computing

Thesis submitted in partial fulfilment of the requirements for the degree of

Master of Technology

in

Computer Science and Engineering

(Specialization: Computer Science)

by

Pranesh Das

Department of Computer Science and Engineering

National Institute of Technology Rourkela

Rourkela – 769 008, India

Virtualization and Fault Tolerance

in Cloud Computing

Dissertation submitted in

Jun 2013

to the department of

Computer Science and Engineering

of


in partial fulfillment of the requirements

for the degree of

Master of Technology

by

Pranesh Das

(Roll 211CS1270)

under the supervision of

Prof. Pabitra Mohan Khilar

Department of Computer Science and Engineering


Rourkela – 769 008, India

Computer Science and EngineeringNational Institute of Technology RourkelaRourkela-769 008, India. www.nitrkl.ac.in

Jun 3, 2013

Certificate

This is to certify that the work in the thesis entitled Virtualization and Fault

Tolerance in Cloud Computing by Pranesh Das, bearing roll number

211CS1270, is a record of an original research work carried out by him under my

supervision and guidance in partial fulfilment of the requirements for the award of

the degree of Master of Technology in Computer Science and Engineering.

Neither this thesis nor any part of it has been submitted for any degree or academic

award elsewhere.

Prof. Pabitra Mohan Khilar

Dept. of Computer Science and Engineering

National Institute of Technology, Rourkela

Acknowledgment

First of all I would like to thank, my Parents, without their moral support

and blessings I wouldn’t have been writing this “thesis”.

I would like to express my deep sense of respect and gratitude towards my

supervisor Prof. Pabitra Mohan Khilar, who has been the guiding force behind

this work. Without his unconditional support it wouldn’t have been possible. As

my supervisor, he has constantly encouraged me to remain focused on achieving my

goal. I want to thank him for giving me the opportunity to work under him. His

invaluable advice and assistance helped me to complete this thesis. I consider it my

good fortune to have got an opportunity to work with such a wonderful person.

I am very much indebted to Prof. Ashok Kumar Turuk, Head-CSE, for his

continuous encouragement and support. He is always ready to help with a smile.

I am also thankful to all the professors of the department for their support. I am

really thankful to all my friends. My sincere thanks to everyone who has provided

me with kind words, a welcome ear, new ideas, useful criticism, or their invaluable

time, I am truly indebted.

I must acknowledge the academic resources that I have got from NIT Rourkela.

I would like to thank administrative and technical members of the Department who

have been kind enough to advise and help in their respective roles.

Last,but not the least, I would like to dedicate this thesis to my family, for their

love, patience, and understanding.

Pranesh Das

Abstract

Fault tolerance in cloud computing is a grand challenge problem now a days. The

main fault tolerance issues in cloud computing are detection and recovery. To

combat with these problems, many fault tolerance techniques have been designed

to reduce the faults. In this research work,a Virtualization and Fault Tolerance

(VFT) technique is used to reduce the service time and to increase the system

availability. A Cloud Manager (CM) module and a Decision Maker (DM) are

used in this scheme to manage the virtualization, load balancing and to handle

the faults. In the first step virtualization and load balancing are done and in

the second step fault tolerance is achieved by redundancy, checkpointing and fault

handler.The main load balancing issues in cloud computing is load calculation and

load distribution. To solve these issues, many load balancing techniques have been

designed to distribute tasks properly. In this work, a Load Balancing Technique

for Virtualization and Fault Tolerance in Cloud Computing (LBVFT) is applied

to assign the tasks to the virtual nodes.LBVFT is mainly designed to assign tasks

to the virtual nodes depending on the success rates (SR) and the previous load

history. In our load assigning technique assignment is done by the load balancer

(LB) of cloud manager (CM) module in the basis of higher success rate and lower

load of the available nodes.A Randomized Searching Algorithm is designed to select

a virtual node.Performance of the Randomized Searching Algorithm lies between

Binary Search and Linear Search Algorithms. VFT is mainly designed to provide

reactive as well as proactive fault tolerance. In this approach a fault handler is

included in the virtualization part. Fault handler blocks the unrecoverable faulty

nodes along with their virtual nodes for future requests and removes the temporary

software faults from the recoverable faulty nodes and makes them available for the

future requests.

Keywords: Fault Tolerance, Cloud Virtualization, Hypervisor, Cloud Computing, Load

Balancing, LBVFT, VFT, Cloud Manager, Decision Maker, Fault handler,Success Rate, Load,

Algorithm, complexity analysis, Binary Search, Linear Search.

Contents

Certificate ii

Acknowledgement iii

Abstract iv

List of Figures viii

List of Tables ix

List of Abbreviations x

1 Introduction 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Virtualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Fault Tolerance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3.1 Reactive Fault Tolerance . . . . . . . . . . . . . . . . . . . . . 3

1.3.2 Proactive Fault Tolerance . . . . . . . . . . . . . . . . . . . . 4

1.4 Load Balancing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.5 Basic Concepts of Cloud Computing . . . . . . . . . . . . . . . . . . 5

1.5.1 Cloud Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.5.2 Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.5.3 Disadvantages . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.6 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

v

1.7 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.8 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Literature Survey 9

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Proposed Virtualization and Fault Tolerance Model(VFT) for

Cloud Computing 11

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.2 Working of the Proposed VFT Model . . . . . . . . . . . . . . . . . . 12

3.2.1 Cloud Manager(CM) . . . . . . . . . . . . . . . . . . . . . . . 12

3.2.2 Load Balancer . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2.3 Fault Handler . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2.4 Decision Maker(DM) . . . . . . . . . . . . . . . . . . . . . . . 14

3.3 Fault Tolerance Technique . . . . . . . . . . . . . . . . . . . . . . . . 14

3.4 Success Rate Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.5 Simulation Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.6 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.7 Conclusion and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Proposed Load Balancing Technique for the VFT Model in Cloud

Computing(LBVFT) 21

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.2 Working of the LBVFT Algorithm . . . . . . . . . . . . . . . . . . . 22

4.3 Different Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.4 Simulation Results and Discussion . . . . . . . . . . . . . . . . . . . . 25

4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

vi

5 Proposed Randomized Searching Algorithm to Select a Node in

the VFT Model 27

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.2 Working of the Proposed Algorithm . . . . . . . . . . . . . . . . . . . 28

5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5.3.1 Performance Analysis of the Proposed Algorithm . . . . . . . 29

5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

6 Conclusion and Scope of Future Work 35

Bibliography 37

Dissemination of Work 43

vii

List of Figures

3.1 Proposed VFT Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.2 Continuous increase in success rate (SR) . . . . . . . . . . . . . . . . 17

3.3 Continuous decrease in success rate (SR) . . . . . . . . . . . . . . . . 17

3.4 Pass to fail shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.5 Fail to pass shifting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.7 Virtual node 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.8 Virtual node 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.9 Virtual node 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.1 No.of operations as a tree structure . . . . . . . . . . . . . . . . . . . 30

5.2 Average Case behaviour of the proposed Algorithm and the existing

Binary Search Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 33

5.3 Average Case Behaviour of the Proposed Algorithm and the existing

Linear Search Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 33

5.4 Average Case Performance Analysis of the Proposed Algorithm,

Binary Search and Linear Search Algorithms. . . . . . . . . . . . . . 33

5.5 Worst Case Behaviour of the Proposed Randomized Search Algorithm

and the existing Binary Search Algorithm. . . . . . . . . . . . . . . . 33

viii

List of Tables

5.1 Simulated Result for Average Case(no.of elements and no.of

comparisons) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5.2 Simulated Result for Worst Case(no.of elements and no.of comparisons) 32

ix

List of Abbreviations

• VFT : Virtualization and Fault Tolerance

• CM: Cloud Manager

• DM: Decision Maker

• LBVFT: Load Balancing Technique for Virtualization and Fault Tolerance

• SR: Success Rate

• LB: Load Balancer

x

Chapter 1

Introduction

1.1 Introduction

Fault tolerance is an approach where a system continues to success even if

there is a fault [2,3]. Although there are number of fault tolerant models

or techniques are available but still fault tolerance in cloud computing is a

challenging task [4,5,6,7,8,9,11]. Because of the very large infrastructure of cloud

and the increasing demand of services an effective fault tolerant technique for

cloud computing is required. In our proposed model fault tolerance is integrated

with the cloud virtualization [22,23]. The basic mechanism to achieve the fault

tolerance is replication or redundancy. We have performed this replication in form

of software variants running on multiple virtual machines. We have presented

a virtualization approach with the help of hypervisor where the load balancer

takes high responsibility by distributing loads only to those virtual nodes whose

corresponding physical servers have a good performance history. To measure the

performance history of a physical server we have used success rate. If n1=number

of times a physical server gives successful results and n2=total number of times

requests sent to that server, then the Success Rate SR =n1/n2, where n1<=n2.

An issue in distributed scheduling is load balancing which tries to distribute the

1

Chapter 1 Introduction

tasks to be executed among the resources of the system [31,32,33]. Load balancing

can be achieved either locally or in a distributed fashion. Distributing tasks across a

communication medium is sometimes referred to as the resource allocation problem.

Resource allocation actually refers to scheduling multiple resources. Here a load

assigning technique LBVFT for the proposed VFT model is proposed. The proposed

load assigning scheme for the VFT model assigns a task to the available virtual

nodes depending on their success rates and the load history. Because of the very

large infrastructure of cloud and the increasing demand of services an effective fault

tolerant technique for cloud computing is required and for which an effective load

balancing approach is required. In the VFT model the load balancer takes high

responsibility by distributing loads only to those virtual nodes whose corresponding

physical servers have a good performance history.

A randomized algorithm is one that receives, in addition to its input data, a stream

of random bits that it can use for the purpose of making random choices. Even for

a fixed input, different runs of a randomized algorithm may give different results

[38,39].In our proposed Randomized Searching Algorithm we choose a position

randomly in each iteration and then the element of that position is compared with

the key item.The performance of the Proposed Algorithm lies between Binary Search

and Linear Search.

Chapter Organization: Section 1.1 describes the introduction,section 1.2

gives a brief view on virtualization,section 1.3 focuses on fault tolerance and its

different types,section 1.4 gives the load balancing concept,section 1.5 basic concepts

of cloud computing,section 1.6 motivation,section 1.7 objectives,section 1.8 thesis

organization and section 1.9 conludes the chapter.

1.2 Virtualization

Virtualization is an emerging IT paradigm that separates computing functions and

technology implementations from physical hardware. Virtualization technology

2


allows servers and storage devices to be shared and utilization be increased.

Applications can be easily migrated from one physical server to another.

1.3 Fault Tolerance

There are various faults which can occur in cloud computing .Based on fault tolerance

policies various fault tolerance techniques can be used that can either be task level

or workflow level[9].

1.3.1 Reactive Fault Tolerance

Reactive fault tolerance policies reduce the effect of failures on application execution

when the failure effectively occurs. There are various techniques which are based on

these policies like Checkpoint/Restart, Replay and Retry and so on.

1. Check pointing/ Restart:

When a task fails, it is allowed to be restarted from the recently checked

pointed state rather than from the beginning. It is an efficient task level fault

tolerance technique for long running applications [2].

2. Replication:

Various task replicas are run on different resources, for the execution to succeed

till the entire replicated task is not crashed. It can be implemented using tools

like HAProxy, Hadoop and AmazonEc2 etc.

3. Job Migration:

During failure of any task, it can be migrated to another machine. This

technique can be implemented by using HAProxy.

4. Task Resubmission:

It is the most widely used fault tolerance technique in current scientific

3


workflow systems. Whenever a failed task is detected, it is resubmitted either

to the same or to a different resource at runtime.

5. User defined exception handling:

In this user specifies the particular treatment of a task failure for workflows.

6. Rescue workflow:

This technique allows the workflow to continue even if the task fails until it

becomes impossible to move forward without catering the failed task.

1.3.2 Proactive Fault Tolerance

The principle of proactive fault tolerance policies is to avoid recovery from faults,

errors and failures by predicting them and proactively replace the suspected

components other working components. Some of the techniques which are based

on these policies are Preemptive migration, Software Rejuvenation etc.

1. Software Rejuvenation:

It is a technique that designs the system for periodic reboots. It restarts the

system with clean state.

2. Proactive Fault Tolerance using Self Healing:

When multiple instances of an application are running on multiple virtual

machines, it automatically handles failure of application instances.

3. Proactive Fault Tolerance using Preemptive Migration:

Preemptive Migration relies on a feedback-loop control mechanism where

application is constantly monitored and analyzed.

1.4 Load Balancing

Load balancing is a computer networking method for distributing workloads across

multiple computers or a computer cluster, network links, central processing units,

4


disk drives, or other resources. Successful load balancing optimizes resource use,

maximizes throughput, minimizes response time, and avoids overload. Using

multiple components with load balancing instead of a single component may increase

reliability through redundancy. Load balancing is usually provided by dedicated

software or hardware, such as a multilayer switch or a Domain Name System server

Process.

1.5 Basic Concepts of Cloud Computing

Cloud computing refers to applications and services that run on a distributed

network using virtualized resources and accessed by common internet protocols and

networking standards.It is distinguished by the notion that resources are virtual and

limitless and that details of the physical systems on which the software runs are

abstracted from the user.

The use of the word ”cloud” makes reference to the two essential concepts:

Abstraction:

Cloud computing abstracts the details of system implementation from users and

developers.Applications run physical systems that are not specified,data is stored

in location that are unknown,adminitration of system are outsourced to others and

access by users is ubiquitous.

Virtualization:

Cloud computing virtualizes systems by pooling and sharing resources.Systems and

storage can be provisioned as needed from a centralized infrastructure,costs are

accessed on a metered basis,multi tenancy is enabled and resources are scalable with

agility.

1.5.1 Cloud Types

Most people separate cloud computing into two distinct sets of models:

5


Deployment models:

This refers to the location and management of the cloud’s

infrastructure.Example-Public cloud,Private cloud,Hybrid cloud and community

cloud.

Service models:

This consists of the particular types of services that we can access on a cloud

computing platform.

• Infrastructure as a service.

• Platform as a service.

• Software as a service.

1.5.2 Advantages

Lower computer costs:

Not necessary to have high-powered computers to access web applications. Even

with cheaper computer also can give efficient results because data is stored in the

web not with us.

Improved performance:

Everything is run in cloud so our computer doesnt have to take much effort to run

applications. As a result, performance will be improved automatically.

Unlimited storage capacity:

Storage is also one kind of service provided by the Cloud, so there is no limit to

store data (based on the service provider).

Device independence:

The actual documents are in the Cloud, so we can access it wherever we are.

6


1.5.3 Disadvantages

Requires a constant High speed Internet connection:

To get benefit from this we need to have always a high speed Internet connection.

Stored data might not be secure:

There is no guarantee that your data stored is in cloud is securely protected.

Intruders may access to your vital data at any time.

1.6 Motivation

After the study, we found that the main fault tolerance issues in cloud computing

are detection and recovery. To combat with these problems, many fault tolerance

techniques have been designed to reduce the faults.But due to its virtualization and

internet based service providing behaviour fault tolerance in cloud computing is still

a big challenge. Our proposed model is not only to tolerate faults but also to reduce

the chance of future faults by not assigning tasks to virtual nodes of physical servers

whose success rates are very low.

1.7 Objective

Our objectives are:

• to design a fault tolerance model which can provide reactive and proactive

fault tolerance in cloud computing.

• to design a load balancing technique for the proposed model to assign tasks to

the virtual nodes.

• to design a searching algorithm to search a virtual node for task assignment.

7


1.8 Thesis Organization

The rest of the thesis is organized as follows: In Chapter 2,we give the literature

review, In Chapter 3, we proposed A Virtualization and Fault Tolerance Model for

Cloud Computing(VFT). In Chapter 4, we proposed A Load Balancing Technique

for Virtualization and Fault Tolerance in Cloud Computing(LBVFT) for the VFT

model. In Chapter 5, we proposed A Randomized Searching Algorithm to select a

node for task assignment. Finally in Chapter 6, we conclude our thesis.

1.9 Conclusion

In this chapter the background concepts for the remaining chapters are given.The

basic concepts of cloud computing,fault tolerance,load balancing are briefly discussed

here.Then the objectives and motivation of the thesis work are discussed.

8

Chapter 2

Literature Survey

2.1 Introduction

In this chapter we briefly discuss the research conducted so far in fault tolerance and

load balancing. There are number of fault tolerance and load balancing algorithms

exist, but they are mostly based on either reactive or proactive fault tolerance but

not both.

2.2 Literature Survey

A lot of work has been done in the area of fault tolerance and load balancing for

cloud computing. But due to its virtualization and internet based service providing

behaviour fault tolerance and load balancing in cloud computing are still a big

challenge. Many researchers have given various fault tolerance techniques and

strategies in [4,5,6,7,8,9,10,11,12,20,21,24,25]. Dilbag Singh and Jaswinder Singh in

[4] have given failover strategies for cloud computing using integrated checkpointing

algorithms. Sheheryar Malik and Fabrice Huet in [10] have given an approach for

adaptive fault tolerance in real time cloud computing.Many researchers have given

various load balancing techniques in [30,31,32,33,34,35,37].The proposed LBVFT

technique distributes loads in a smart way by considering the success rates and the

9

Chapter 2 Literature Survey

loads history of the available virtual nodes. Thus the LBVFT helps the VFT model

to tolerate not only faults but also reduce the chance of future faults by not assigning

tasks to virtual nodes of physical servers whose success rates are very low and loads

are very high.

Chapter Organization Section 1.1 gives the introduction,section 1.2 related

work or literature survey is discussed and section 1.3 concludes the chapter.

2.3 Conclusion

After the analysis and survey on the related works and some selected research works

on fault tolerance in cloud computing,load balancing and virtualization it is seen that

fault tolerance in cloud computing is a challenging task.This thesis work proposes a

model for fault tolerance.

10

Chapter 3

Proposed Virtualization and Fault

Tolerance Model(VFT) for

Cloud Computing

3.1 Introduction

The Virtualization and Fault Tolerance (VFT) model is proposed here which pro-

vides the reactive fault tolerance on cloud infrastructure. This scheme tolerates the

faults on the basis of Success Rate (SR) (0 < SR <= 1) of each virtual node’s phys-

ical server. A virtual node is selected for computation on the basis of SR of its

corresponding physical server and can be removed, if the selected nodes physical

server does not perform well. Our model consists of two main modules Cloud

Manager (CM) module and Decision Maker (DM) module .The model is shown

in figure 1.

Chapter Organization: In this chapter section 1.1 introduces the

chapter,section 1.2 working of the proposed model,section 1.3 fault tolerance

technique,and section 1.4 success rate analysis of the nodes are given.Section 1.5

simulation result,section 1.6 evaluation of the simulated results and section 1.7

11

Chapter 3Proposed Virtualization and Fault Tolerance Model(VFT) for

Cloud Computing

concludes the chapter.

Figure 3.1: Proposed VFT Model

3.2 Working of the Proposed VFT Model

In our proposed scenario a client submits a task to the Cloud Service Provider

(CSP).The CSP then submits it to the Cloud Manager (CM).

3.2.1 Cloud Manager(CM)

Cloud Manager is included in the cloud architecture. It performs the virtualization

with the help of Hypervisor. Hypervisor is a low level program which creates a

virtual environment and provides system resource access to virtual machines. When

virtual nodes are created from the available resources of the physical servers (System

Hardware) then Hypervisor maintains a record of which virtual node belongs to

which physical node. Resources of a single physical server can be used to create set

12


Cloud Computing

of virtual nodes. A Performance Record (PR) table is maintained containing the

servers ids, virtual nodes ids and Success Rate (SR) to identify the virtual nodes

and to keep record of the number of times tasks are assigned to the virtual nodes

of a particular server and gets successful result from those virtual nodes of the

corresponding physical server. Here the SR is the success rate of a physical node

whose resources are used for virtualization. The PR table is also available to the

Load Balancer.

3.2.2 Load Balancer

Load Balancer distributes the loads based on the Performance Record (PR) of the

physical systems that are used for virtualization .Load Balancer assigns tasks to

the virtual nodes with the help of load balancing algorithms[26], [27], [28]. The

Load Balancer always assigns the tasks to those virtual nodes whose corresponding

physical servers are having good SR.

3.2.3 Fault Handler

Fault Handler takes the responsibility when a virtual node is found faulty due to

some recoverable temporary software faults in the CM or due to some transient

faults occurred in the remote physical server of the corresponding virtual node .The

PR table gets updated .If there is no virtual node executing under that physical

server then Fault Handler immediately restart the remote server and informs the

Load balancer not to assign any task to the virtual nodes of the corresponding

server .The Fault Handler may apply some transient fault detection and recovery

technique[4],[5].If the fault handling is done successfully then the virtual node or

virtual nodes of that physical server are made available for future request.Results

that are executed by the virtual nodes are submitted to the Decision Maker (DM)

module.

13


Cloud Computing

3.2.4 Decision Maker(DM)

Decision Maker is also included in the cloud architecture like CM. Status Checker

(SC) in DM checks the status of each virtual node. If status is success then Task

Deadline is checked by TDC sub module in DM. If both Status Checking and Task

Deadline checking (TDC) is success then SR of the corresponding node is increased

and it goes for Final Decision Making (FDM) sub module. If either SC or TDC

is fail then the corresponding virtual machine is not sent to the FDM sub module.

If the SC is fail then the node is sent to the Fault Handler for fault detection and

recovery. If SC is success but the task is not completed within the time limit then

also the node is not considered for final decision making and SR in the PR table

is decreased for that node. FDM sub module contains all the virtual nodes that

successfully executed the SC and TDC modules. FDM then selects a node with the

highest SR value and makes a checkpoint [13], [14], [15], [16], [29]. If more than one

node is having same SR value then selects any one randomly. If all nodes are failed

then backward recovery is performed with the help of the last successful checkpoint.

3.3 Fault Tolerance Technique

In our proposed VFT model we apply SR computation algorithm for each node

(virtual machine) one by one. Initially SR of a node is set to 0.5. This

algorithm takes input of two factors, PR table (SR, n1, n2, vmid; serverid)and

maxSuccessRate.maxSuccesRate is the maximum SR level. Node SR cannot be

more than this level. In our proposed scheme maxSuccessRate is 1. It is really

important in a situation , where an initially produces correct results in consecutive

cycles, but then fails again and again. In our proposed scheme the SR value cannot

be zero because Load Balancer will not assign any task to a virtual node whose

corresponding server nodes SR is less than equal to zero.

14


Cloud Computing

Success Rate Computation Algorithm:

1. Initially success rate =0.5, n1=1, n2=2

2. n1 is the number of times the virtual node of a particular physical server gives

successful results

3. n2 is the number of times the Load Balancer of the cloud manager(CM) assigns

tasks to a particular servers virtual node

4. Input maxSuccessRate=1

5. Status of a node is Success if SC and TDC module for that node is success

6. Status of a node is Fail if SC or TDC or both module for that node is fail

7. if (nodeStatus = =Success) /*SC and TDC success*/

{

n1=n1+1

n2=n2+1

SuccessRate = n1/n2

Update PR table

}

8. else

{

if( nodeStatus = =Fail ) /* SC or TDC or both fail */

{

n2=n2+1

SuccessRate= n1/n2

Update PR table

}

}

15


Cloud Computing

9. if (SuccessRate>=maxSuccessRate)

{

SuccessRate = maxSuccessRate

}

10. if(SuccessRate<=0)

{

Reject the node and inform the Cloud Manager to add a new node

}

Decision Technique Algorithm:

1. Initially SuccessRate=0.5

2. Input from TDC:nodeSuccessRate, n=number of nodes with SC and TDC

success

3. Input maxSuccessRate

4. if(n= =0)

{

Status = fail

Perform backward recovery with the last successful checkpoint

}

5. else

{

Status =Success

bestSuccessRate = find SuccessRate of node with highest SuccessRate

Select the node with bestSuccessRate and send the result to CSP

Make checkpoint

}

16


Cloud Computing

3.4 Success Rate Analysis

Here we have given a metric analysis to analyse the success rate for different scenarios

of four virtual nodes. We have taken 100 computing cycles. In the beginning

we have assumed that success rate is 0.5. In figure 3.2, figure 3.3, figure 3.4 and

figure 3.5, a comparison analysis is given between success and failure scenarios.

This comparison is done for 100 computing cycles. In these cycles vm1 continuously

succeeded, vm2 continuously failed, vm3 succeeded for first 50 cycles and then failed

for remaining 50 cycles and vm4 failed for first 50 cycles and then succeeded for

remaining 50 cycles. The increase in success rate after 100 cycles for vm1 is 0.8545,

whereas decrease in success rate for vm2 is 0.0545, and increase in success rate for

vm3 and vm4 is 0.4.We have assumed that each virtual node belongs to a different

physical server. Here we can see that increase in success rate is more than decrease.

Hence we can achieve a good performance of the algorithm.

Figure 3.2: Continuous increase in success

rate (SR)

Figure 3.3: Continuous decrease in success

rate (SR)

17


Cloud Computing

Figure 3.4: Pass to fail shifting Figure 3.5: Fail to pass shifting

3.5 Simulation Result

We have done the simulation in Cloudsim 2.0 [1], [18], [19] with NetBeans IDE 6.7.1.

In this simulation we have created three virtual nodes. Every virtual node executes

the same task at a time. The algorithm has 10 computing cycles. The algorithms

consist of a series of tasks. Each of these tasks is performed in one computing cycle.

Each virtual node may run diverse algorithm.

Figure 3.6:

Then we have DM module which is receiving the results from the virtual machines

18


Cloud Computing

to be processed. This DM is integrated with the CM at the Cloud Service Provider

(CSP) [1] site. If a node fails then the system will continued to be executed using

the remaining nodes. The system continues its execution until all nodes are failed.

After a successful computing cycle a node is selected and a checkpoint is made by

the FDM to keep the status of the system for future recovery. Here we have assumed

that the following information of the selected virtual nodes is available to us: PR

table i.e. value of n1, n2, SR, virtual node id and corresponding server id. Initially

n1=1 and n2=2 i.e. SR=0.5 is considered for every node. Tasks deadlines are also

given as input. The simulated results are shown on Figure 3.6 and from those results

success rate analysis of node1, node2 and node3 is shown on Figure 3.7, Figure 3.8

and Figure 3.9 respectively.

Figure 3.7: Virtual node 1 Figure 3.8: Virtual node 2

Figure 3.9: Virtual node 3

19


Cloud Computing

3.6 Evaluation

In the first cycle, all nodes have the same success rate, so any one can be selected.

We have selected node 1.In cycle 2 node 2 is selected. In most of the remaining

cycles node 1 is selected as its SR is increases continuously. In cycle 9 virtual node2

did not pass SC and so TDC also failed and node 1 is selected as it has the highest

SR. If SC is fail then TDC must be fail , e.g. cycle 9. But if SC passes then TDC

may pass or may not. In case of either SC and TDC failure or only SC failure an

error message is sent to the fault handler of the CM module and to the TDC sub

module. The error handler then tries to repair the fault. TDC received it before

time limit. But as there is no result produced, so the status of TDC is also fail. In

Figure 3.7, Figure 3.8 and Figure 3.9 success rate(SR)fluctuation is shown.

3.7 Conclusion and Discussion

This paper proposes a smart failover strategy for cloud computing using success

rate of the computing nodes and virtualization which include the support of load

balancing algorithms and fault handler. A quantitative analysis is given in section

V. Performance comparison of existing methods has been made with the purposed

method. It has been concluded with the help of performance metrics comparison

and success rate analysis from simulated results that the proposed fault tolerant

strategy gives a very good performance. In our future work we will work on the

Fault Handler and Load Balancer sub modules of CM module in order to make the

model more fault tolerant.

20

Chapter 4

Proposed Load Balancing

Technique for the VFT Model in

Cloud

Computing(LBVFT)

4.1 Introduction

Here in the proposed load balancing approach when cloud manager (CM) of VFT

model gets the request from the cloud service provider (CSP) it creates set of virtual

nodes with the help of system hardware, host operating system and hypervisor

module. System hardware means set of physical servers which are connected by

distributed network .CM then gives the responsibility to the load balancer to assign

the tasks. As in the VFT model to achieve fault tolerance redundant virtual nodes

are used so, the LB will assign the same task to a number of available virtual nodes

which are having good SR value and lower load in the performance record table.

Thus the same task will be executed in all the selected nodes and the result will be

sent to the Decision Maker (DM) module of the VFT model (figure 1).

21

Chapter 4Proposed Load Balancing Technique for the VFT Model in Cloud

Computing(LBVFT)

Chapter organization: Section 1.1 introduces the chapter,section 1.2 provides

the working of the proposed load balancing technique,section 1.3 different scenarios

of the algorithm,section 1.4 gives the simulation result and section 1.5 concludes the

chapter.

4.2 Working of the LBVFT Algorithm

1. Initially success rate (SR)=0.5,maximum SR=1, 0 < SR <= 1.

2. Input high SR, higher SR, low SR, lower SR, low load, lower load, high load

and higher load values.

3. SR=n1/n2

4. n1 is the number of times the virtual node of a particular physical server gives

successful results.

5. n2 is the number of times the Load Balancer of the cloud manager(CM) assigns

tasks to a particular servers virtual node.

6. Loads of the virtual nodes are given. Loads are generally calculated by the

load balancer and is updated and kept in the performance record table of the

virtual nodes time to time, which is beyond the scope of this paper.

7. Search for all the available virtual nodes having lower load history and higher

SR values in the performance record table.

8. if((SR == higher||SR == high)&&(load == lower))

{

Select the node

}

9. else

{

22


Computing(LBVFT)

if((SR == higher||SR == high)&&(load == low))

{

Select the node

}

}

10. if((SR == Higher||SR == High)&&(load == higher))

{

Dont select the node if enough nodes are

available

}

11. if((SR == Higher||SR == High)&&(load == high))

{

Dont select the node if enough nodes are

available

}

12. if ((SR == low||SR == lower)&&(load == low))

{

Select the node if numbers of nodes are less

}

13. if ((SR == low||SR == lower)&&(load == lower))

{

Select the node if numbers of nodes are less

}

14. if((SR == low||SR == lower)&&(load == high))

{

23


Computing(LBVFT)

Dont select the node

}

15. if((SR == low||SR == lower)&&(load == higher))

{

Dont select the node

}

16. Select m nodes from the list of available virtual nodes (m is the user input;

m-1 is the number of redundant nodes).

17. Submit the task to all the selected m nodes.

18. The tasks are then executed by the allocated virtual nodes and the results are

sent to the decision maker (DM) module of the VFT model.

4.3 Different Scenarios

1. High or higher success rate and low or lower load: In this case a node

is better one for selection and the node is selected. As the load is low and the

SR value is high, there is a less chance of future fault.

2. High or higher success rate and high or higher load: If a node having

this property is selected then there is a chance of future fault as the load is

high. So, this kind of nodes is avoided for selection if sufficient numbers of

virtual nodes are available.

3. Low or lower success rate and high or higher load A node having this

feature is not selected, because the SR value is low as well as load is high.

4. Low or lower success rate and low or lower load In this case a node may

be selected if sufficient numbers of nodes are not available.

24


Computing(LBVFT)

4.4 Simulation Results and Discussion

Simulation is done in Cloudsim2.0 [9,10] with NetBeans IDE 6.7.1.The simulation

result that is obtained from VFT model [1] is given in Table 4.4.1.In the resultant

table L1, L2.L33 are the loads of the virtual nodes at that instant of time.

Figure 4.1:

From Table 4.4.1 another table Table 4.4.2 is constructed. It is assumed that

there are 3 virtual nodes available and their SR values and loads are taken .Here

it is considered that there are 10 tasks and for every task 3 virtual nodes are

there.Proposed load assigning technique will assign tasks to those nodes which have

good SR and load values. It is assumed that number of redundant node is 1.Hence

for every task 2 virtual nodes will be selected by the proposed technique which is

shown in Table 4.4.2.

25


Computing(LBVFT)

Figure 4.2:

4.5 Conclusion

This paper gives a smart load distribution strategy for our virtualization and fault

tolerance in cloud computing (VFT) model using success rate of the computing

nodes and previous load history. In Table 4.4.2 it is seen that maximum time node

1 and 2 is selected. For task t8 and task t4 although node 1 and node 2 has high

SR values still are not selected because it is assumed that loads for those nodes

are very high. This task assigning technique helps the VFT model to give a good

performance. As only higher SR values and lower loads are considered during virtual

node selection hence, there is a very less chance of system failure. Our future work

is to provide an efficient load balancing, load migration, load calculation and fault

handling technique to make the VFT model more effective.

26

Chapter 5

Proposed Randomized Searching

Algorithm to Select a Node in

the VFT Model

5.1 Introduction

In our proposed Randomized Searching Algorithm the main idea is to select positions

randomly from the sorted input array and compare elements of those positions with

the search key until the match is found or the array is finished. If the key element is

found in the randomly selected position p then we stop. If the search key is less than

or greater than the element of the chosen position then the searching is continued

in the left side or in the right sight of the position respectively. We have done the

simulation in MATLAB 7.12.0(R2011a).We have generated the elements of the input

array A randomly. We also have generated the search key element K randomly and

done the simulation for a large no of times with arrays of different sizes.For a given

load or SR value a particular node can be searched in the VFT model using this

algorithm only if the performance record table is sorted according to load history or

success rate.

27

Chapter 5Proposed Randomized Searching Algorithm to Select a Node in

the VFT Model

Chapter Organization: In chapter section 1.1 gives the introduction,section

1.2 working of the proposed algorithm,section 1.3 gives the results and performance

analysis of the algorithm and section 1.4 concludes the chapter.

5.2 Working of the Proposed Algorithm

Begin:

1. lb=lower bound of the sorted input array

2. ub=upper bound of the input array

3. while( lb<=ub )

{

p=choose a number randomly between lb and ub

if( K<A(p))

{

ub=p-1

}

else

if(K>A(p))

{

lb=p+1

}

else

{

return p

}

}

4. return −1 /* couldn’t not find the key */

28


the VFT Model

End

5.3 Results

5.3.1 Performance Analysis of the Proposed Algorithm

Complexity Analysis:

We see that in the first two lines of the working of the proposed RSA before the

while loop, 2 primitive operations always get executed (two assignments). However,

since these operations happen no matter what the input is, we will ignore them for

now. Focusing our attention on the while loop, we see that each time the program

enters the while loop, we execute 3+k primitive operations (a <= comparison, k

number of operations to chose a random number between lb and ub, an array index,

and a < comparison), before the program might branch depending on the result of

the first if statement.

Depending on the result of the conditional, the program will execute different

numbers of primitive operations. If key > A[p], then the program executes 2 more

primitive operations. If key < A[p], then the program executes 4 more primitive

operations (an array index and a < comparison in the next if, and then a subtraction

and assignment). If A[p] = = key, then we execute 3 more primitive operations (2

operations for the if and then 1 operation to return position p).As k is fix for each

iteration so we neglect k now. In other words, the number of primitive operations

executed in an iteration of the loop is

5, if key < A[p]

6 , if A[p] = = key, and

7 , if key > A[P].

We can now construct a tree that summarizes how many operations are executed in

the while loop, depending on the number of times the while loop is executed and

the result of the comparisons between A[p] and key in each iteration. Each node in

29


the VFT Model

the tree represents the exit from the algorithm because A[p] = = key. A node is a

left child of its parent if in the previous iteration of the while loop, the search was

restricted to the left part of the subarray (that is, when A[P] > key). A node is a

right child of its parent if in the previous iteration of the while loop, the search was

restricted to the right part of the sub array (that is, when A[p] < key).

Here are the first three levels of the tree: In case of Binary Search algorithm height

Figure 5.1: No.of operations as a tree

structure

of the left sub tree and the height of the right sub tree are equal and we can easily

proof that the average case time complexity for binary search is log n.Because the

number of nodes are equal in both the left and right sub trees and total number of

nodes on each level can be easily computed.But in our proposed algorithm as the

position is randomly chosen so the height of left sub tree and the height of right sub

tree may not be equal.

Hence to analyse the average case behaviour we have done a simulation on our

proposed algorithm. We have done the simulation on the basis of no.of element

versus no.of comparisons instead of number operations.In our described algorithm

on section 0.4.2 the relevant number of comparisons(one <= comparison,one < and

one >) executed in an iteration of the loop is:

30


the VFT Model

2, if key < A[p]

3 , if A[p] = = key, and

3 , if key > A[P].

We have simulated the proposed algorithm in MATLAB 7.12.0(R2011a).In the

simulation we have generated the elements of the arrays randomly and also the

key.The simulation is done for 50 times with different number of elements ranges

from 10 to 500.The simulated result for average case is shown below:

Table 5.1: Simulated Result for Average Case(no.of elements and no.of comparisons)

No.of Compar No.of Compari No.of Compari No.of Compari No.of Compari

elements isons elements sons elements sons elements sons elements sons

10 7 110 14 210 24 310 22 410 19

20 14 120 18 220 17 320 29 420 22

30 20 130 18 230 30 330 22 430 19

40 12 140 24 240 41 340 25 440 16

50 24 150 25 250 18 350 28 450 24

60 8 160 32 260 21 360 19 460 32

70 12 170 20 270 22 370 33 470 23

80 33 180 23 280 15 380 30 480 29

90 26 190 23 290 18 390 30 490 39

100 25 200 28 300 19 400 15 500 25

From the above table Table 1 we see that in average approximately 3 l0g n (base 2)

comparisons are required to find the key item for the different set of elements.Hence

we can say that approximately the time complexity of our proposed Randomized

Searching Algorithm is O(3 log n) in Average Case.

31


the VFT Model

For Best Case every time we find the key in the first attempt i.e in the first

loop and only 3 comparison are performed for every set of elements.

In Worst Case key item is not in the input array i.e in every iteration

maximum number of comparisons are performed.The simulation is done in MATLAB

7.12.0(R2011a) for the worst case analysis.Here also we have generated the keys and

the elements randomly and done the simulation for 50 times with different number

of elements ranges from 10 to 500.The simulated result for Worst Case is shown

below:

Table 5.2: Simulated Result for Worst Case(no.of elements and no.of comparisons)

No.of Compar No.of Compari No.of Compari No.of Compari No.of Compari

elements isons elements sons elements sons elements sons elements sons

10 9 110 27 210 31 310 21 410 45

20 18 120 20 220 23 320 24 420 21

30 10 130 14 230 14 330 26 430 26

40 18 140 20 240 20 340 21 440 24

50 16 150 10 250 25 350 20 450 31

60 27 160 22 260 26 360 38 460 21

70 16 170 19 270 26 370 28 470 27

80 18 180 32 280 22 380 23 480 12

90 20 190 13 290 29 390 15 490 18

100 15 200 14 300 25 400 15 500 28

From the above table Table 2 we observe that approximately the same number of

comparisons are required to find the key items for the different set of elements as

like as the average case.Hence we can say that in Worst Case also approximately the

time complexity is O(3 log n).

32


the VFT Model

Efficiency Graphs Compared to Binary Search and Linear Search

Algorithms:

Figure 5.2: Average Case behaviour of

the proposed Algorithm and the existing

Binary Search Algorithm

Figure 5.3: Average Case Behaviour of

the Proposed Algorithm and the existing

Linear Search Algorithm

Figure 5.4: Average Case Performance

Analysis of the Proposed Algorithm,

Binary Search and Linear Search

Algorithms.

Figure 5.5: Worst Case Behaviour of the

Proposed Randomized Search Algorithm

and the existing Binary Search Algorithm.

33


the VFT Model

Here on the above figures we have shown the performance analysis of our proposed

algorithm graphically. In figure 5.2 we see that our Proposed Algorithm is performing

approximately same as Binary Search in average case. During simulation it is seen

that some times our proposed algorithm gives better result then the Binary Search.

Here we did not get smooth graph because of randomization. From figure 5.3 we see

that our proposed algorithm is giving better result then the existing linear Search

Algorithm (although linear search is applied on unsorted array).

In figure 5.4 average case behaviour of Linear Search, Binary Search and RSA is

shown. In figure 5.5 worst case analysis is done. It is seen that average case and

worst behaviour are nearly same for our Proposed RSA Algorithm.

5.4 Conclusion

Performance of the proposed RSA algorithm lies between Binary Search and

Linear Search Algorithms. With respect to Linear Search our algorithm gives

better performance although Linear Search is applied on unsorted data. Due to

randomization we can get an average result every time. We will improve the

performance of the algorithm by designing a proper algorithm to choose a number

randomly.

34

Chapter 6

Conclusion and Scope of Future

Work

This paper proposes a smart failover strategy for cloud computing using success

rate of the computing nodes and virtualization which include the support of load

balancing algorithms and fault handler. A quantitative analysis is given in section

V. Performance comparison of existing methods has been made with the purposed

method. It has been concluded with the help of performance metrics comparison

and success rate analysis from simulated results that the proposed fault tolerant

strategy gives a very good performance. In our future work we will work on the

Fault Handler and Load Balancer sub modules of CM module in order to make the

model more fault tolerant. This paper gives a smart load distribution strategy for

our virtualization and fault tolerance in cloud computing (VFT) model using success

rate of the computing nodes and previous load history. In Table 2 it is seen that

maximum time node 1 and 2 is selected. For task t8 and task t4 although node 1

and node 2 has high SR values still are not selected because we assumed that loads

for those nodes are very high.This task assigning technique helps the VFT model to

give a good performance. As only higher SR values and lower loads are considered

during virtual node selection hence, there is a very less chance of system failure. Our

35

Conclusion and Scope of Future Work

future work is to provide an efficient load balancing, load migration, load calculation

and fault handling technique to make the VFT model more effective. Performance

of the proposed RSA algorithm lies between Binary Search and 41 Linear Search

Algorithms. With respect to Linear Search our algorithm gives better performance

although Linear Search is applied on unsorted data. Due to randomization we can

get an average result every time. We will improve the performance of the algorithm

by designing a proper algorithm to choose a number randomly.

36

Bibliography

[1] http://www.cloudbus.org/cloudsim/

[2] Barrie Sosinsky, “Cloud Computing ”, Wiley India Edition, ISBN:

978-81-265-2980-3.

[3] Reese, G., “Cloud Application Architectures: Building Applications and

Infrastructure in the cloud (Theory in Practice) ”, OReilly Media, 1st Ed.,

2009 pp 30-46.

[4] Dilbag Singh, Jaswinder Singh, Amit Chhabra, “High Availability of Clouds:

Failover Strategies for Cloud Computing using Integrated Checkpointing

Algorithms”, IEEE International Conference on Communication Systems and

Network Technologies, 2012.

[5] G. A. Gibson, B. Schroeder, and J. Digney, “Failure Tolerance in Petascale

Computers”, CTWatchQuarterly, vol 3,no 4, November 2007.

[6] Ifeanyi P. Egwutuoha, Shiping Chen, David Levy, Bran Selic, “A Fault

Tolerance Framework for High Performance Computing in Cloud”, 12th

IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing,

2012.

[7] Ghassem Miremadi, Johan Karlsson, Ulf Gunnejlo, Jan Torin, “Two

Software Techniques for On-line Error Detection”, Twenty-Second

37

Bibliography

International Symposium on Computing Processing (Hardware/Software),

10.1109/FTCS.1992.243622.

[8] Dimitris Gizopoulos, Mihalis Psarakis, Sarita V. Adve, Pradeep Ramachandran,

Siva Kumar Sastry Hari, Daniel Sorin, Albert Meixner, Arijit Biswas, and

Xavier Vera, “Architectures for Online Error Detection and Recovery in

Multicore Processors”, Proceedings of the Design, Automation and Test in

Europe (DATE), March 2011.

[9] Anju Bala, Inderveer Chana , “Fault Tolerance- Challenges, Techniques

and Implementation in Cloud Computing”, IJCSI International Journal of

Computer Science Issues, vol. 9, Issue 1, no 1, January 2012.

[10] Sheheryar Malik, Fabrice Huet, Adaptive Fault Tolerance in Real Time Cloud

Computing”, IEEE World Congress on Services, 2011.

[11] Wenbing Zhao, P. M. Melliar-Smith and L. E. Moser, “Fault Tolerance

Middleware for Cloud Computing”, IEEE 3rd International Conference on

Cloud Computing, 2010.

[12] Ekpe Okorafor, “A Fault-tolerant High Performance Cloud Strategy for

Scientific Computing”, IEEE International Parallel Distributed Processing

Symposium, 2011.

[13] Sangho Yi and Derrick Kondo, Artur Andrzejak, “Reducing Costs of Spot

Instances via Checkpointing in the Amazon Elastic Compute Cloud”, IEEE

3rd International Conference on Cloud Computing, 2010.

[14] Dilbag Singh, Jaswinder Singh, Amit Chhabra, “Evaluating Overheads

of Integrated Multilevel Checkpointing Algorithms in Cloud Computing

Environment”, I. J. Computer Network and Information Security, 2012, 5,

29-38.

38

Bibliography

[15] Sangho Yi, Member, IEEE, Artur Andrzejak, and Derrick Kondo, Member,

IEEE, “Monetary Cost-Aware Checkpointing and Migration on Amazon Cloud

Spot Instances”, IEEE Transactions on Services Computing, vol. x, no. x,

month 201X.

[16] Jose Carlos Sancho Fabrizio Petrini Greg Johnson Juan Fernandez Eitan

Frachtenberg, “On the Feasibility of Incremental Checkpointing for Scientific

Computing”, Proceedings of the 18th International Parallel and Distributed

Processing Symposium (IPDPS04), 2004, IEEE.

[17] Jianjian Song1, Heng Kek Choo1 and Kuok Ming Lee2, “Application-level load

migration and its implementation on top of PVM”, Concurrency: Practice and

Experience, vol. 9(1), 119 (January 1997).

[18] Rodrigo N. Calheiros, Rajiv Ranjan, Anton Beloglazov, Csar A. F. De Rose,

and Rajkumar Buyya, “CloudSim: A Toolkit for Modeling and Simulation

of Cloud Computing Environments and Evaluation of Resource Provisioning

Algorithms”, Software: Practice 2011 - Wiley Online Library .

[19] Rajkumar Buyya, Rajiv Ranjan and Rodrigo N. Calheiros, “Modeling and

Simulation of Scalable Cloud Computing Environments and the CloudSim

Toolkit: Challenges and Opportunities”, International Conference on High

Performance Computing Simulation, 2009. HPCS ’09,P-1-11.

[20] Andreas Menychtas, Kleopatra G. Konstanteli, “Fault Detection and Recovery

Mechanisms and Techniques for Service Oriented Infrastructures”, IGI Global,

2012.

[21] Yi Hu, Bin Gong, Fengyu Wang, “Cloud Model-Based Security-aware and

Fault-Tolerant Job Scheduling for Computing Grid”, The Fifth Annual

ChinaGrid Conference, IEEE, 2010.

39

Bibliography

[22] Anthony T. Velte, Toby J. Velte, Robert Elsenpeter, “Cloud Computing: A

Practical Approach”, the McGraw-Hill Companies, ISBN: 978-0-07-162695-8,

MHID: 0-07-162695-6, 2010.

[23] Qingling Wang, Carlos A. Varela, Impact of Cloud Computing Virtualization

Strategies on Workloads Performance”, Fourth IEEE International Conference

on Utility and Cloud Computing (UCC), 5-8 Dec. 2011,P-130-137.

[24] N. Yadav and P.M.Khilar, “Hierarchically Adaptive Distributed Fault Diagnosis

in Mobile Adhoc Networks using Clustering”, in proc. of International

Conference on Industrial and Information System 2010 (ICIIS 2010), NITK,

Surathkal, Karnatak, India during 29th July to 1stAugust 2010.

[25] Pabitra Mohan Khilar, Jitendra Kumar Singh, Sudipta Mahapatra, “Design

and Evaluation of a Failure Detection Algorithm for Large Scale Ad Hoc

Networks Using Cluster Based Approach”, IEEE International Conference on

Information Technology,10.1109/ICIT.2008.72.

[26] Martin Randles, Enas Odat, David Lamb, Osama Abu- Rahmeh and A.

Taleb-Bendiab, “A Comparative Experiment in Distributed Load Balancing”,

Second International Conference on Developments in Systems Engineering

,10.1109/DeSE.2009.20,P-258- 265.

[27] Martin Randles, David Lamb, A. Taleb-Bendiab, “A Comparative Study

into Distributed Load Balancing Algorithms for Cloud Computing”, IEEE

24th International Conference on Advanced Information Networking and

Applications Workshops (WAINA), 2010, P: 551- 556.

[28] A.Khiyaita , M.Zbakh, H. El Bakkali ,Dafir El Kettani , Load Balancing Cloud

Computing : State of Art”, National Days of Network Security and Systems

(JNS2), 2012, 10.1109/JNS2.2012.6249253, P: 106- 109.

40

Bibliography

[29] Guohong Cao, Student Member, IEEE, and Mukesh Singhal, Member, IEEE,

“On Coordinated Checkpointing in Distributed Systems”, IEEE Transactions

on Parallel and Distributed System, Vol. 9, No. 12, December 1998.

[30] Neeraj Rathore, Dr. Inderveer Chana , “A Cognitive Analysis of Load Balancing

and job migration Technique in Grid”, World Congress on Information and

Communication Technologies (WICT), 2011,P 77-82, ISBN 978-1-4673-0127-5.

[31] A.K, M.Z, EL Bakkali, Dafir EL Kettani, “Load Balancing Cloud Computing

: State of Art”, ISBN 978-1-4673-1053-6 , IEEE 2012.

[32] Jinhua Hu ,Jianhua Gu, Guofei Sun, Tianhai Zhao, “A Scheduling Strategy

on Load Balancing of Virtual Machine Resources in Cloud Computing

”, 3rd International Symposium on Parallel Architectures, Algorithms and

Programming , 978-0-7695-4312-3/10, 2010 IEEE,DOI 10.1109/PAAP.2010.65.

[33] Xiaona Ren , Rongheng Lin, Hua Zou, “A Dynamic Load Balancing Strategy

for Cloud Computing Platform Based on Exponential Smoothing Forecast”,

Proceedings of IEEE CCIS2011 , 978-1-61284-204-2/11, 2011 IEEE.

[34] M. Randles, D. Lamb, and A. Taleb-Bendiab, “A Comparative Study

into Distributed Load Balancing Algorithms for Cloud Computing”, 2010

IEEE 24th International Conference on Advanced Information Networking and

Applications Workshops , 2010, pp. 551556.

[35] Derek L. Eager, Edward D. Lazowska, Jhon Zahorjan, “Adaptive Load Sharing

in Homogeneous Distributed ”, IEEE Transactions on Software Engineering ,

Vol. SE-12,No 5,May 1986.

[36] Rodrigo N. Calheiros, Rajiv Ranjan, Anton Beloglazov, Csar A. F. De Rose,

and Rajkumar Buyya, “CloudSim: A Toolkit for Modeling and Simulation

of Cloud Computing Environments and Evaluation of Resource Provisioning

Algorithms”, Software: Practice 2011 - Wiley Online Library.

41

Bibliography

[37] Rajkumar Buyya, Rajiv Ranjan and Rodrigo N. Calheiros, “Modeling and

Simulation of Scalable Cloud Computing Environments and the CloudSim

Toolkit: Challenges and Opportunities”, International Conference on High

Performance Computing Simulation, 2009. HPCS ’09,P-1-11.

[38] Francesc J. Ferri and Enrique Vidal, “Case-Studies on Average-Case Analysis

for an Elementary Course on Algorithms”, IEEE TRANSACTIONS ON

EDUCATION,VOL. 42, NO. 2, MAY 1999.

[39] Karp, R.M., Introduction to randomized algorithms,Discrete Applied

Mathematics”, Computer Science Division,University of California

Berkeley,International Computer Science Institute,Berkeley, CA 94704, USA

.

42

Dissemination of Work

1. Pranesh Das and Pabitra Mohan Khilar, VFT: A Virtualization and Fault

Tolerance Approach for Cloud Computing˝, IEEE International Conference

on Information and Communication Technology(ICT 2013), Norul Islam

University,Kanyakumari,Tamilnadu, April 11-12, 2013. (In press)

2. Pranesh Das and Pabitra Mohan Khilar,LBVFT: A Load Balancing

Technique for Virtualization and Fault Tolerance in Cloud Computing

˝,International Journal of Computer Application(IJCA),Vol 69,Issue 28,Page

14-18,2013.(Published)

3. Pranesh Das and Pabitra Mohan Khilar, A Randomized Searching

Algorithm and its Performance Analysis with Binary Search and Linear

Search Algorithms ˝, The International Journal of Computer Science and

Application(TIJCSA), Vol 1,No 11, 2013. (Published))

43

virtualization and fault tolerance in cloud...

Documents